Details are provided about a hybrid mapping strategy where a map is constructed from the robot self-movements. The expression of noise in the robot trajectory can be such that different space descriptions are extracted even when the robot traverses the environment repeatedly in the same way. This has a negative impact on the graph representing the environment, which would have many paths representing the same portion of the environment. This increase in map redundancy is obviated by introducing in the robot the capacity to forget the portions of the map that are uncommon. Two such mechanisms are proposed and tested in a simulated Khepera. Their impact on the map is assessed by evaluating some features of the resulting graph; one in particular is a measure of how stable is the map, which we call energy